Hi, all,
How to lower tensor to matrix instructions, like 16*16 matrix/tensor multiply or add, rather than two nests loops with operations in tvm?
Thanks!
this is a planned feature that is yet to be officially announced so please stay tuned
I'm interested in this too. Do you have any rough timeline on developing this feature or the required infrastructure?
All the elements are in, it is mainly effort of documentation and testing
that's great!
close this for now and will update when more documents get in
Most helpful comment
this is a planned feature that is yet to be officially announced so please stay tuned